Constraint-Based Multi-agent Path Planning

Planning collision-free paths for multiple robots traversing a shared space is a problem that grows combinatorially with the number of robots. The naive centralised approach soon becomes intractable for even a moderate number of robots. Decentralised approaches, such as priority planning, are much faster but lack completeness. Previous work has demonstrated that the search can be signifreduced by adding a level of abstraction (Ryan 2008). I first partition the map into subgraphs of particular known structures, such as cliques, halls and rings, and then build abstract plans which describe the transitions of robots between the subgraphs. These plans are constrained by the structural es of the subgraphs used. When an abstract plan is found, it can easy be resolved into a complete concrete plan without further search. In this paper, I show how this method of planning can be implemented as a constraint satisfaction problem (CSP). Constraint propagation and intelligent search ordering furth...

• Type: Conference
• Year: 2008
• Where: AUSAI
• Authors: Malcolm Ryan
• Keywords: Abstract Plan, Artificial Intelligence, AUSAI 2008, Multiple Robots, Naive Centralised Approach